Jamming detection at the edge of drone networks using Multi-layer Perceptrons and Decision Trees

Abstract

As wireless networks play an increasingly key role in everyday life, it is necessary to secure them from radio frequency attacks, such as jamming, which are hard to detect, especially because they may be easily mistaken for other network conditions. Within this challenging context, the paper proposes a framework for jamming detection in drone networks, relying on a distributed approach based on supervised machine learning techniques, namely, Multi-layer Perceptrons and Decision Trees. Given a reference data packet trace set, our framework computes the features of some predefined metrics, such as throughput, PDR and RSSI, which vary during a jamming attack, and that can therefore be used to detect it. We evaluate our framework using datasets from publicly available standardized jamming attack scenarios with IEEE 802.11p radio data, and via ns3-based simulation datasets from networks of drones using WiFi. We show that the performance of the classifiers improves as the sampling time of the packets decreases. We also show that the Multi-layer Perceptron can be effectively generalized to achieve jamming detection accuracy superior to that of Decision Trees even when applied to communication scenarios for which it has not been specifically trained. Our proposed framework reaches a satisfactory accuracy level of 96%, while requiring low computational and hardware capabilities, thus proving to be suitable for resource-constrained drone networks.